# Can Price Variations Predict Trade Flow Volumes?

*Reading time: 8 minutes*

*This study showcases how AgFlow’s API helps you quickly extract all sorts of agricultural commodities’ prices & import-export data to build robust analytical trading models.*

Since 2013, we have been building AgFlow’s database, which currently contains close to 5 million data points today, including:

- Cash price quotes
- Trade flows
- Freight indications
- Tenders

AgFlow not only has the most extensive database in terms of agricultural commodities price data – it’s also the most granular available out there. For the vast majority of our cash prices, we provide four individual quotes:

- Buyer price
- Seller price
- Nominal price
- Traded price

And if you were to search for trade flows volumes of Corn coming out of Brazil and going to Vietnam, you would find see vessel name, status, date sailed, cargo volume, export and import ports, and charterers.

The wealth of data on cash prices and trade flows volumes left us wondering:

**What if spot prices of specific products were correlated to trade flow volumes?**

**And if so, can we use price variations to predict changes in trade flow volumes?**

## Methodology

**Step 1:** We extracted price and trade flow volumes data for four commodities

- Argentina Soybean Oil (Soyoil) FOB
- Brazil Corn FOB
- Indonesia Palm Oil FOB
- Ukraine Wheat FOB

**Step 2:** Sum of all monthly trade flows volumes for each product

**Step 3:** Price aggregation

**Step 4:** Shipments and forward prices analysis at respectively one, two, and three months

**Step 5:** Correlations between price and shipment periods

**Step 6:** Jointplots to simultaneously visualize prices, trade flows volumes, and their respective distribution

**Step 7:** Results analytics each separate shipment window

## Argentina Soyoil FOB

First, let’s look at shipments occurring one month after a quotation.

Interestingly enough, we witnessed a low correlation between trade flows volumes and quotes average prices for Argentina Soyoil FOB shipments in the linear regression fitting, which is counter-intuitive. To evaluate this fit quantitatively, we applied the R-squared function (R²) that measures the regression fit to the data.

Source: AgFlow.com

**Figure 1: R² Values for Argentina Soyoil FOB, Shipment = 1 month**

Source: AgFlow.com

**Figure 2: Jointplots for Argentina Soyoil FOB and Shipment Windows at 1 month **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

However, when looking at the polynomial fitting, the correlation coefficient grows stronger. This suggests a dependency on more than the two variables, namely – price and trade flow volumes. However, since the additional variables remain unknown to this point, the level of incertitude increases, and this is why the incertitude channel grows wider in figure 2 and reaches almost the whole range of prices for the data at the extremities for the third-degree polynomial fit, represented in figure 3.

Therefore, only considering price variations to predict trade flow volumes changes for Argentina Soyoil, does not provide enough rationale to build a sound trading strategy. However, we can conclude that additional unknown factors that must be identified are playing a role in driving Argentina Soyoil FOB trade flow volumes changes.

In order to pinpoint said variables, a good starting point would be to analyze datapoints standing out of the distribution and research why they occurred. For instance, was this event driven by a change in weather, by logistics complications, or by regulation constraints?

Looking at Figure 1 a little closer, we were stricken by a few datapoints sticking out to the distribution. While we would have usually excluded them, we decided against it because they represent real market events, and therefore we cannot consider them as outliers.

The point highlighted in the red diamond in the graph above corresponds, for instance, to shipment in May 2018. Proceeding similarly for shipments at 2 and 3 months, and looking specifically for the highest outstanding prices, they both corresponded to June 2018. We can, therefore, understand that a specific event during that short period has created a market move. Was it a change in import or export tax? Arbitrage? A logistic event? Once the data behind the event is identified, it can be built in the analysis as a new factor and test for prediction performance movement.

Additionally, the position of the points in the volumes axis is also interesting. These shipments volumes are located in the lower end of the spectrum. It’s, therefore, safe to assume that since there is a lower quantity shipped at that period there must be, according to rules of supply-and-demand, high demand at that time for the product.

Source: AgFlow.com

**Figure 3: Jointplot for Argentina Soyoil FOB and Shipment Windows at 1 month, ****Linear Fit**

Now, looking at the shipment periods occurring two and three months after the quotation and their respective R² values, we observe that the further the prices are from the shipments months, the stronger the correlation coefficient grows.

Source: AgFlow.com

**Figure 4: Jointplots for Argentina Soyoil FOB and Shipment Windows at 2 month **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

Source: AgFlow.com

**Figure 5: Jointplots for Argentina Soyoil FOB and Shipment Windows at 3 month **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

Source: AgFlow.com

**Figure 6: R² Values for Argentina Soyoil FOB, Shipment = 2 & 3 months**

In conclusion,

First, results indicate that predictions are not only driven by two variables – price and trade flow volumes. There are more variables – which remain unknown to this point – that are driving trade flow volumes variation and quotes prices variations.

Second, we also demonstrated that fitting higher polynomial models, while improving prediction performance, also increases the width of the incertitude. Therefore, it’s critical to consider the R² values of models as well as what level of incertitude areas they bring.

Third, we’ve observed that the aptest prediction is at three months from the quotation i.e., for a price formulated three months before shipment. These prices are also the closest and most correlated ones with regard to shipments.

In a nutshell, our analysis shows that there is little to no correlation between trade flows volumes and quotes prices for Argentina Soyoil FOB and that higher polynomial fits suggest that additional variables need to be included in the model to improve its prediction performance.

Having performed similar processing on Brazil Corn, Indonesian Palm Oil, and Ukrainian Wheat. We found that Brazil Corn is better fitted for this type of analysis.

Read Also: Coronavirus Impact Soybean Meal Prices

## Brazil Corn FOB

Source: AgFlow.com

**Figure 7: Jointplots for Brazil Corn FOB and Shipment Windows at 1, 2, 3 and months **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

Source: AgFlow.com

**Figure 8: Jointplots for Brazil Corn FOB and Shipment Windows at 2 months **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

Source: AgFlow.com

**Figure 9: Jointplots for Brazil Corn FOB and Shipment Windows at 3 months **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

Source: AgFlow.com

**Figure 10: R² Values for Brazil Corn FOB, Shipment = 1, 2 & 3 months**

First, it can easily be spotted that the Brazil Corn FOB distribution does not follow the same pattern as Argentina’s Soyoil FOB.

Comparing both distributions for Argentina Soyoil FOB and Brazil Corn FOB prices for Brazil Corn FOB follow a rather normal distribution and fit the Gaussian curve quite well. Inversely, the Argentina Soyoil FOB distribution curve has a lesser normal distribution.

Moreover, we can logically see the outstanding values that were observed previously fit to the data is much better overall, making it a stronger candidate for prediction.

In addition to this, the correlation is degrading as the shipment timeframe grows wider. And the case of Brazil Corn FOB, it is when the shipment is equal to a month after quotation that it is the best fit and the most relevant. If we take a closer look at the R² for this shipment window, as well as observing the incertitude, we can see that once more the higher polynomials logically have a better fit to the data and that the incertitudes channels are also increasing with polynomial degrees.

In this case, though, the difference between R² values is smaller, yet the increase in incertitude still is significant (though smaller than for Argentina Soyoil FOB). Therefore it can be understood that whilst the higher polynomial models are interesting, they don’t necessarily describe the data very well.

In conclusion, Brazil Corn FOB is probably less dependent on other variables, and as such, that all commodities don’t necessarily follow the same rules.

Source: AgFlow.com

**Figure 11: Distribution Plots Argentina Soyoil FOB**

Source: AgFlow.com

**Figure 12: Distribution Plots Brazil Corn FOB**

Looking at Indonesia Palm Oil FOB and Ukraine Wheat FOB, confirms that there is no general rule that applies to all commodities and that all have their own specificities.

Source: AgFlow.com

**Figure 13: R² Values for Argentina Soyoil FOB, Shipment = 1 month**

## Indonesia Palm Oil FOB

The Indonesian Palm Oil FOB distribution is very similar to the Argentinean Soyoil FOB in terms of volume distribution but presents a better regression fit as the difference in prices is smaller than for Argentina Soyoil FOB. As such it strengthens the idea that the correlation between both values depends strongly on more variables

Source: AgFlow.com

**Figure 14: Jointplots for Indonesia Palm Oil FOB and Shipment Windows at 1 month **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

## Ukraine Wheat FOB

Ukraine Wheat FOB does not follow the same rules as the 3 previous distributions. Price values are much closer to each other there appears to be any variation with trade flow volumes. Furthermore, looking at shipment windows, it appears that most particularly the last 2 and 3 months timeframes have strikingly similar data distribution patterns, incertitude areas, and R² values for their respective polynomial fits.

In addition to observing that, once more a different product obeys different rules, we can see that shipment periods that are further than a month have very limited influence on the correlation between the Volumes and Prices. Furthermore, it informs on the volatile aspect of the market as well. With very small variations of prices depending on volume or time, it shows that the product is potentially very stable, perhaps due to stable supply-and-demand.

Source: AgFlow.com

**Figure 15: Jointplots for Ukraine Wheat FOB and Shipment Windows at 1 month **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

Source: AgFlow.com

**Figure 16: Jointplots for Ukraine Wheat FOB and Shipment Windows at 2 months **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

Source: AgFlow.com

**Figure 17: Jointplots for Ukraine Wheat FOB and Shipment Windows at 3 months **(1. Linear Fit, 2. Second Degree Polynomial Fit, 3. Third Degree Polynomial Fit)

Source: AgFlow.com

**Figure 18: R² Values for Ukraine Wheat FOB, Shipment = 1, 2 & 3 months**

Read Also: Coronavirus, Wheat & Food Inflation

## Conclusions

We were wondering whether or not we could use the correlation between quotes prices and the trade flow volumes at different shipment windows; to predict trade flows volumes and their prices.

We’ve learned that not only the correlations were not strong enough between both variables to use them to formulate any meaningful prediction but that these relationships were also dependent on other potential variables such as production, meteorological conditions, logistics, or regulation changes.

Moreover, we demonstrated that not all products were following the same rules nor had the same dependencies.

This study was done using a limited amount of data in a short time frame, and yet still yielded great insights into the complex relationships of price quotes and trade flows in the agricultural commodity trading space.

It could also be interesting to further study this question by extending shipment quotation months or compare similar commodities from different origins, or even study how the correlations of an origin with trade flow volumes evolve across commodities.